- Aug 8, 2025
- 61
- 96
- 69
- Thread starter Bulik98
- Start date
- Aug 8, 2025
- 61
- 96
- 69
From a programmer standpoint (and as someone that's been integrating AI with IT software solutions outside of this niche), I highly agree with everyone about the issue with the AI chat, depending on the LLM model (especially if it's API-based) it could cost the developer a hefty amount.
to break it down in simpler terms, an API depending on both the model (like Claude 4.5, etc) and provider (OpenAi/Alibaba/Google Gemini)generally has a per "million" token pricing. Add to the complexity that boolah probably needs to find a decent LLM suitable for NSFW roleplaying, even the cheapest or probably an extremely lightweight model would be around 0.4$ per million tokens, which could be considered cheap at first, but considering that the average length of a player's input token could range around 100-400 tokens per prompt (depends on the context mostly) - which is already around 0.00004$ - 0.00016$ per prompt (on an individual), say for example that per individual here could be as much as 20 prompts (for extremely lightweight users or players)
that would be estimated around 0.0008$ - 0.0032$ per individual, and assume including the paid and free users use that for free without any daily, or even monthly limits (the problem is that there's no account registration so boolah probably has no way to track and monitor that), Now let's assume that daily there's at least 50 individual users with that kind of token consumption, that would be like 0.04$ - 0.16$ per day, and let's assume that's stable for a whole month, the monthly estimated expenses would be 1.2$(30 * 0.04$) - 4.8$(30 * 0.16$).
and that's just for the really lightweight LLM, extremely low amount of prompts and individual users. By standard a decent roleplaying LLM is somewhere around 1$-2$ per million of token(input + output),
that would roughly cost around 3$ - $12 even with the lightweight consumption of tokens and minimal users.
Now if we roughly try to guess that the LLM boolah uses costs around that price, then factor in that boolah probably had way more than 50 users daily for the AI, let's say 200 users for that kind of token consumption (just a really really really really rough guess even with just 20 prompts average per user) - the cost immediately shoots through the roof with around 12$ - 48$ per month.
from a marketing/business perspective, there's not even a break-even or even a net profit to at least maintain the ai chat, as it gives extremely low profit per user per average consumption... and on average, the expenses starts to outweigh the remaining gross income the more boolah's fanbase probably increases
to break it down in simpler terms, an API depending on both the model (like Claude 4.5, etc) and provider (OpenAi/Alibaba/Google Gemini)generally has a per "million" token pricing. Add to the complexity that boolah probably needs to find a decent LLM suitable for NSFW roleplaying, even the cheapest or probably an extremely lightweight model would be around 0.4$ per million tokens, which could be considered cheap at first, but considering that the average length of a player's input token could range around 100-400 tokens per prompt (depends on the context mostly) - which is already around 0.00004$ - 0.00016$ per prompt (on an individual), say for example that per individual here could be as much as 20 prompts (for extremely lightweight users or players)
that would be estimated around 0.0008$ - 0.0032$ per individual, and assume including the paid and free users use that for free without any daily, or even monthly limits (the problem is that there's no account registration so boolah probably has no way to track and monitor that), Now let's assume that daily there's at least 50 individual users with that kind of token consumption, that would be like 0.04$ - 0.16$ per day, and let's assume that's stable for a whole month, the monthly estimated expenses would be 1.2$(30 * 0.04$) - 4.8$(30 * 0.16$).
and that's just for the really lightweight LLM, extremely low amount of prompts and individual users. By standard a decent roleplaying LLM is somewhere around 1$-2$ per million of token(input + output),
that would roughly cost around 3$ - $12 even with the lightweight consumption of tokens and minimal users.
Now if we roughly try to guess that the LLM boolah uses costs around that price, then factor in that boolah probably had way more than 50 users daily for the AI, let's say 200 users for that kind of token consumption (just a really really really really rough guess even with just 20 prompts average per user) - the cost immediately shoots through the roof with around 12$ - 48$ per month.
from a marketing/business perspective, there's not even a break-even or even a net profit to at least maintain the ai chat, as it gives extremely low profit per user per average consumption... and on average, the expenses starts to outweigh the remaining gross income the more boolah's fanbase probably increases
Last edited:
- Aug 8, 2025
- 61
- 96
- 69
to top off the extreme expenses, there's no login or account management so there's really no way for boolah to track and limit usage per user even if it gets pirated, and because there's no user monitoring, there's probably no data analytics for boolah's side. and from what I know the ai chat isn't really optimized to compress or save token consumptions as reported by some users with the ai forgetting some context making token consumption even higher as the story progresses more...
Last edited:
wolfund
Active Member
- Aug 22, 2017
- 627
- 810
- 350
We never said it was THE cause for it. Simply that when it come down to it was a business mistake as it was a ressource sink in every way. And that this fact more than likely factored in when the real and more pressing reason of Boola having a temper tantrum over having a hard time dealing with external pressure over the game direction reared its head.
The AI chat problem is that it's money and time burnt for something of little value in game. It would have been better spent making the LLM model and selling THAT to one of those many site out there doing just that.
Illustrivity
Member
- Mar 2, 2018
- 367
- 953
- 316
If the AI chat feature was really the reason this game was abandoned, then that might be the most absolutely dumbshit reason for a game to get abandoned of all time. I genuinely never gave a shit about the feature, and apparently neither did most other people. What an absolute pointless waste of time.
wolfund
Active Member
- Aug 22, 2017
- 627
- 810
- 350
The most likely MAIN reason was the pressure and negativity over the direction of the game. The AI chat might have weighted in too but it's mostly a minor reason and a business mistake.
I mean the last update pissed off a lot of people it could be that he abandoned his project because of that.
I had a very long and detailed conversation with Boola, also about his AI chat. He always had problems with the AI because the storylines became too complex, and he didn't know how to implement the individual plot threads so that they would make sense when users chatted with Mia. He was always looking for an AI expert to help him; I think the cost was only a small part of why he gave up. He always bought 1000 questions per update, so each paying user had a limit of 50 questions, and free users only 5. That was the basic idea. Of course, the fact that his chat was quickly cracked is also a factor, as there were too many users in the chat who hadn't actually paid for it. I once saw a graph of chat users and the volume of questions; it was considerable, I think 14,000 questions in one day. Then Boola limited everything because the chat crashed and nothing worked anymore. Boola clearly overestimated himself here. As the game progressed, the chat became unmanageable for him, let alone implementing it in a way that allowed Mia to remember everything that happened in the game. So I don't agree that the chat was worthless, as it was based on the gameplay. His time management also suffered greatly; he couldn't release updates as planned because the chat was too buggy and time-consuming. This was very evident in his last release, where he completely botched it.
- Jun 6, 2020
- 663
- 4,850
- 477
I believe there was DeepSeek API in source code
Truly a dumb decision if there really was paid LLM chat embedded, Pruned DeepSeek model could be easily installed with the game making users run it offline for free
I wonder if free local AI chat bots are popular niche right now, maybe I should create one...
Honestly after talking to Boolah, and a few other devs I really notice that many of these devs don't understand the complications of game design nor do they understand how to write a long form storyline that branches. In boolah's case the conversations I had were constantly me giving criticism and compliments which was followed by Boolah freaking out and not understanding what was being said. The fact that Boolah was writing a game with branching stories and what was billed as a positive NTS route... yet openly said he didn't understand why anyone would like a non humiliation based NTS story was more than enough to tell me he wasn't prepared to write this game long term. He seemed to regularly struggle to understand basic genre questions about the game. Struggled immensely with critiques and hyper manageing the monthly downloads of the updates within a day or so and asking why people weren't downloading the game that month....during holidays. It was just definitely too much for the poor guy especially when people who had been defending him to the last a few weeks to a month prior became rabid angry hornballs that were tearing apart his mistakes.
wolfund
Active Member
- Aug 22, 2017
- 627
- 810
- 350
To be fair this is very much a "first time exposing your creation" problem. Every budding creator get through that at first. I've been around writing on the net for over fifteen years now, so my skin pretty thick, but the first year or so? Oh man I walked away off project more than I care to remember because of that.
There's just a point where as creator you start to... just go over it. There's a point where critics are easy to discern from attacks disguised as such and your eyes somewhat glaze over the idiots, but like you need actual real creating experience for that. Managing that is a whole over can of worms that a LOT of people just aren't ready for.
It's a pain in the ass, but if anyone of you end up starting making something and put it on the internet, you better be ready for people to come in and do NOTHING but tear it down piece by piece, sometime because there's a genuine reason, sometime just because they have an opportunity to hurt you for shit and giggles.
edit: I always hated the "get a thicker skin" thing when I was younger, but there's really not much else to do. Creating things is a personal, emotional endeavor, any attack on THAT is going to hit deeper than a fist to the face if you're not prepared for it. And there's not much else to do to get a thicker skin than to go on trial by fire every single darn day you put yourself out there.
- Aug 8, 2025
- 61
- 96
- 69
It might be boola's first time integrating AI, and just like what I've seen and guessed, he didn't implemented any prompt engineering or at least token optimization techniques.
if boola didn't abandoned this game or at least gave a heads-up, a lot of us and even i included may have assisted with the issues, and that's if he's open minded enough and willing to learn. As for the individual plot threads, several optimization techniques such as "chunking" or a mini built-in memory system module for the AI, or a general memory map wherein the AI does not always have to be fed with the full context of the chat every user prompt would have saved a lot of tokens and improved performance. a common technique that most AI engineers also use is having or saving a summarized version of the context every certain period of interval (maybe per chapter would do??), which is what will be fed to the AI and depending on the integration, further optimization could have a conditional "lookup" for specific fragments of the chat context if needed so it helps balance out the quality and token efficiency of the AI
- Aug 8, 2025
- 61
- 96
- 69
that would explain a lot of the issues if it is really a DeepSeek LLM. From what I know DeepSeek models are one of the most expensive ones compared to other LLM providers.
I also highly agree that having a local LLM would have a better profitability in the long-run, but I think there are several trade-offs to be considered on boola's side depending on his approach, if we assume boola integrates it in every single installable platforms, he's probably going to run into a lot of "compatibility" and "hardware" issues which would probably impact the performance and portability to most of low-end devices.. on the other hand if he were to host his own local LLM (and probably create his own API key), the country I come from would kind off cost an astronomical amount of funds to get a decent dedicated hardware for training and hosting a personal LLM (that's excluding extremely light weight models such as 7B parameter types, but doing that would just defeat the purpose of creating a LLM in the first place if the AI's not intelligent enough)
Last edited:
- Aug 8, 2025
- 61
- 96
- 69
+1 on this one. I went through a lot of roasting on my "UI" at my initial launch of the game as it was also my first experience in the field of the gaming industry, and in this specifc "niche" as well lol, and still do up to this point. I think one of the best mindset that a game dev should adopt is indeed getting a thicker skin, but at the same time be open-minded enough to take constructive criticisms, as it is what will help us become better on what we do in the long run. I owe a lot as well to the community that gave me a lot of feedbacks. It's just really unfortunate the project got dropped without any heads-up, and I'm not sure how much he reached out to his supporters, but I hoped he at least pm'ed people as well that can help him
This is why I look at the AI art boom in the industry as a bad direction. The benefit of actually learning art traditionally is critique. I went to art school for game design.....it was brutal. Everything you make gets torn apart by your teachers and classmates. but that builds your skin up. You become able to deal with it or you quit in a safe environment. I've had people buy my artwork as a side gig to my full time job. its much easier to be ready for art criticism and pressure. People in the normal world aren't exactly prepared for the pressure of an art timeline and the majority aren't ready for the criticism. AI art people are already going to get crushed by anti AI art critiques which really have no counterargument that is going to save face for them. So they are starting with a massive amount of criticism and when their fangirls in the discords start piling on its just a miserable experience for them. they usually build an echo chamber in their discords and try to ignore critiques until it floods into that chamber and whether they swim or drown is up to them.
I tend to want to point to traditional game coding methods as the best option for maintaining storylines and processes in a game. the issue for anyone using AI as a shortcut in the processes for this is that you are relying on a system you don't have a full understanding of. there's a reason most big companies build their own game engines for their big games. they created it so they know everything it can do. And if they need it to do something new they can add it.
The big problem for Boolah is that he wasn't exactly prepared for the game he was developing. He didn't understand the coding, writing, or the systems he was putting in. He started very well with a basic phone game that had a great series of small additions like the system of photos having mystery clues you could click on to have the MC notice shit in photos and change dialogue which was abandoned way too soon. or the AI texting thing which was not abandoned soon enough.
- Aug 8, 2025
- 61
- 96
- 69
In the professional world we refer to those "traditional game coding methods" as native programmers. It's not limited to just in the gaming industry, but generally in any area that involves "programming". I think a lot of people, and not just a lot but most people that are just looking for "shortcuts" are using AI in the wrong way that doesn't provide proper growth or learning.
just like what you've mentioned, I think most of people that use AI as a shortcut alone for games, probably do not have a formal or professional background/education in programming. from what I can see, boolah might have been one of those. The problem with relying too much with AI, especially in the domain of programming is that some people do not have an understanding why a certain bug happens (even the most basic ones), or how a system architecture works (probably no data structure and algorithm/back-end foundation). I've seen quite a bit of games that are generally a mess of spaghetti code and logic that do not make sense that are probably generated by AI alone.
one of the hallmark of a source code that isn't even at least modified or natively coded by a programmer but by an AI is a codebase with entirely no separation of concern like MVC frameworks (like web dev frameworks)